Hi,
I finally got around to reviving this series. However, I simplified it
from v2 and focussed on solving the btrfs search_ioctl() problem only:
https://lore.kernel.org/r/[email protected]
The btrfs search_ioctl() function can potentially live-lock on arm64
with MTE enabled due to a fault_in_writeable() + copy_to_user_nofault()
unbounded loop. The uaccess can fault in the middle of a page (MTE tag
check fault) even if a prior fault_in_writeable() successfully wrote to
the beginning of that page. The btrfs loop always restarts the fault-in
loop from the beginning of the user buffer, hence the live-lock.
The series introduces fault_in_subpage_writeable() together with the
arm64 probing counterpart and the btrfs fix.
I don't think with the current kernel anything other than btrfs
search_ioctl() can live-lock. The buffered file I/O can already cope
with current fault_in_*() + copy_*_user() loops (the uaccess makes
progress). Direct I/O either goes via GUP + kernel mapping access (and
memcpy() can't fault) or, if the user buffer is not PAGE aligned, it may
fall back to buffered I/O. So we really only care about
fault_in_writeable(), hence this simplified series. If at some point
we'll need to address other places, we can expand the sub-page probing
to the other fault_in_*() functions.
Thanks.
Catalin Marinas (3):
mm: Add fault_in_subpage_writeable() to probe at sub-page granularity
arm64: Add support for user sub-page fault probing
btrfs: Avoid live-lock in search_ioctl() on hardware with sub-page
faults
arch/Kconfig | 7 +++++++
arch/arm64/Kconfig | 1 +
arch/arm64/include/asm/mte.h | 1 +
arch/arm64/include/asm/uaccess.h | 15 +++++++++++++++
arch/arm64/kernel/mte.c | 30 ++++++++++++++++++++++++++++++
fs/btrfs/ioctl.c | 7 ++++++-
include/linux/pagemap.h | 1 +
include/linux/uaccess.h | 22 ++++++++++++++++++++++
mm/gup.c | 29 +++++++++++++++++++++++++++++
9 files changed, 112 insertions(+), 1 deletion(-)
With MTE, even if the pte allows an access, a mismatched tag somewhere
within a page can still cause a fault. Select ARCH_HAS_SUBPAGE_FAULTS if
MTE is enabled and implement the probe_subpage_writeable() function.
Note that get_user() is sufficient for the writeable MTE check since the
same tag mismatch fault would be triggered by a read. The caller of
probe_subpage_writeable() will need to check the pte permissions
(put_user, GUP).
Signed-off-by: Catalin Marinas <[email protected]>
Cc: Will Deacon <[email protected]>
---
arch/arm64/Kconfig | 1 +
arch/arm64/include/asm/mte.h | 1 +
arch/arm64/include/asm/uaccess.h | 15 +++++++++++++++
arch/arm64/kernel/mte.c | 30 ++++++++++++++++++++++++++++++
4 files changed, 47 insertions(+)
diff --git a/arch/arm64/Kconfig b/arch/arm64/Kconfig
index 57c4c995965f..290b88238103 100644
--- a/arch/arm64/Kconfig
+++ b/arch/arm64/Kconfig
@@ -1871,6 +1871,7 @@ config ARM64_MTE
depends on AS_HAS_LSE_ATOMICS
# Required for tag checking in the uaccess routines
depends on ARM64_PAN
+ select ARCH_HAS_SUBPAGE_FAULTS
select ARCH_USES_HIGH_VMA_FLAGS
help
Memory Tagging (part of the ARMv8.5 Extensions) provides
diff --git a/arch/arm64/include/asm/mte.h b/arch/arm64/include/asm/mte.h
index adcb937342f1..aa523591a44e 100644
--- a/arch/arm64/include/asm/mte.h
+++ b/arch/arm64/include/asm/mte.h
@@ -47,6 +47,7 @@ long set_mte_ctrl(struct task_struct *task, unsigned long arg);
long get_mte_ctrl(struct task_struct *task);
int mte_ptrace_copy_tags(struct task_struct *child, long request,
unsigned long addr, unsigned long data);
+size_t mte_probe_user_range(const char __user *uaddr, size_t size);
#else /* CONFIG_ARM64_MTE */
diff --git a/arch/arm64/include/asm/uaccess.h b/arch/arm64/include/asm/uaccess.h
index e8dce0cc5eaa..6677aa7e9993 100644
--- a/arch/arm64/include/asm/uaccess.h
+++ b/arch/arm64/include/asm/uaccess.h
@@ -460,4 +460,19 @@ static inline int __copy_from_user_flushcache(void *dst, const void __user *src,
}
#endif
+#ifdef CONFIG_ARCH_HAS_SUBPAGE_FAULTS
+
+/*
+ * Return 0 on success, the number of bytes not probed otherwise.
+ */
+static inline size_t probe_subpage_writeable(const void __user *uaddr,
+ size_t size)
+{
+ if (!system_supports_mte())
+ return 0;
+ return mte_probe_user_range(uaddr, size);
+}
+
+#endif /* CONFIG_ARCH_HAS_SUBPAGE_FAULTS */
+
#endif /* __ASM_UACCESS_H */
diff --git a/arch/arm64/kernel/mte.c b/arch/arm64/kernel/mte.c
index 78b3e0f8e997..35697a09926f 100644
--- a/arch/arm64/kernel/mte.c
+++ b/arch/arm64/kernel/mte.c
@@ -15,6 +15,7 @@
#include <linux/swapops.h>
#include <linux/thread_info.h>
#include <linux/types.h>
+#include <linux/uaccess.h>
#include <linux/uio.h>
#include <asm/barrier.h>
@@ -543,3 +544,32 @@ static int register_mte_tcf_preferred_sysctl(void)
return 0;
}
subsys_initcall(register_mte_tcf_preferred_sysctl);
+
+/*
+ * Return 0 on success, the number of bytes not probed otherwise.
+ */
+size_t mte_probe_user_range(const char __user *uaddr, size_t size)
+{
+ const char __user *end = uaddr + size;
+ int err = 0;
+ char val;
+
+ __raw_get_user(val, uaddr, err);
+ if (err)
+ return size;
+
+ uaddr = PTR_ALIGN(uaddr, MTE_GRANULE_SIZE);
+ while (uaddr < end) {
+ /*
+ * A read is sufficient for mte, the caller should have probed
+ * for the pte write permission if required.
+ */
+ __raw_get_user(val, uaddr, err);
+ if (err)
+ return end - uaddr;
+ uaddr += MTE_GRANULE_SIZE;
+ }
+ (void)val;
+
+ return 0;
+}
On hardware with features like arm64 MTE or SPARC ADI, an access fault
can be triggered at sub-page granularity. Depending on how the
fault_in_writeable() function is used, the caller can get into a
live-lock by continuously retrying the fault-in on an address different
from the one where the uaccess failed.
In the majority of cases progress is ensured by the following
conditions:
1. copy_to_user_nofault() guarantees at least one byte access if the
user address is not faulting.
2. The fault_in_writeable() loop is resumed from the first address that
could not be accessed by copy_to_user_nofault().
If the loop iteration is restarted from an earlier (initial) point, the
loop is repeated with the same conditions and it would live-lock.
Introduce an arch-specific probe_subpage_writeable() and call it from
the newly added fault_in_subpage_writeable() function. The arch code
with sub-page faults will have to implement the specific probing
functionality.
Note that no other fault_in_subpage_*() functions are added since they
have no callers currently susceptible to a live-lock.
Signed-off-by: Catalin Marinas <[email protected]>
Cc: Andrew Morton <[email protected]>
---
arch/Kconfig | 7 +++++++
include/linux/pagemap.h | 1 +
include/linux/uaccess.h | 22 ++++++++++++++++++++++
mm/gup.c | 29 +++++++++++++++++++++++++++++
4 files changed, 59 insertions(+)
diff --git a/arch/Kconfig b/arch/Kconfig
index 29b0167c088b..b34032279926 100644
--- a/arch/Kconfig
+++ b/arch/Kconfig
@@ -24,6 +24,13 @@ config KEXEC_ELF
config HAVE_IMA_KEXEC
bool
+config ARCH_HAS_SUBPAGE_FAULTS
+ bool
+ help
+ Select if the architecture can check permissions at sub-page
+ granularity (e.g. arm64 MTE). The probe_user_*() functions
+ must be implemented.
+
config HOTPLUG_SMT
bool
diff --git a/include/linux/pagemap.h b/include/linux/pagemap.h
index 993994cd943a..6165283bdb6f 100644
--- a/include/linux/pagemap.h
+++ b/include/linux/pagemap.h
@@ -1046,6 +1046,7 @@ void folio_add_wait_queue(struct folio *folio, wait_queue_entry_t *waiter);
* Fault in userspace address range.
*/
size_t fault_in_writeable(char __user *uaddr, size_t size);
+size_t fault_in_subpage_writeable(char __user *uaddr, size_t size);
size_t fault_in_safe_writeable(const char __user *uaddr, size_t size);
size_t fault_in_readable(const char __user *uaddr, size_t size);
diff --git a/include/linux/uaccess.h b/include/linux/uaccess.h
index 546179418ffa..8bbb2dabac19 100644
--- a/include/linux/uaccess.h
+++ b/include/linux/uaccess.h
@@ -231,6 +231,28 @@ static inline bool pagefault_disabled(void)
*/
#define faulthandler_disabled() (pagefault_disabled() || in_atomic())
+#ifndef CONFIG_ARCH_HAS_SUBPAGE_FAULTS
+
+/**
+ * probe_subpage_writeable: probe the user range for write faults at sub-page
+ * granularity (e.g. arm64 MTE)
+ * @uaddr: start of address range
+ * @size: size of address range
+ *
+ * Returns 0 on success, the number of bytes not probed on fault.
+ *
+ * It is expected that the caller checked for the write permission of each
+ * page in the range either by put_user() or GUP. The architecture port can
+ * implement a more efficient get_user() probing if the same sub-page faults
+ * are triggered by either a read or a write.
+ */
+static inline size_t probe_subpage_writeable(void __user *uaddr, size_t size)
+{
+ return 0;
+}
+
+#endif /* CONFIG_ARCH_HAS_SUBPAGE_FAULTS */
+
#ifndef ARCH_HAS_NOCACHE_UACCESS
static inline __must_check unsigned long
diff --git a/mm/gup.c b/mm/gup.c
index f598a037eb04..501bc150792c 100644
--- a/mm/gup.c
+++ b/mm/gup.c
@@ -1648,6 +1648,35 @@ size_t fault_in_writeable(char __user *uaddr, size_t size)
}
EXPORT_SYMBOL(fault_in_writeable);
+/**
+ * fault_in_subpage_writeable - fault in an address range for writing
+ * @uaddr: start of address range
+ * @size: size of address range
+ *
+ * Fault in a user address range for writing while checking for permissions at
+ * sub-page granularity (e.g. arm64 MTE). This function should be used when
+ * the caller cannot guarantee forward progress of a copy_to_user() loop.
+ *
+ * Returns the number of bytes not faulted in (like copy_to_user() and
+ * copy_from_user()).
+ */
+size_t fault_in_subpage_writeable(char __user *uaddr, size_t size)
+{
+ size_t faulted_in;
+
+ /*
+ * Attempt faulting in at page granularity first for page table
+ * permission checking. The arch-specific probe_subpage_writeable()
+ * functions may not check for this.
+ */
+ faulted_in = size - fault_in_writeable(uaddr, size);
+ if (faulted_in)
+ faulted_in -= probe_subpage_writeable(uaddr, faulted_in);
+
+ return size - faulted_in;
+}
+EXPORT_SYMBOL(fault_in_subpage_writeable);
+
/*
* fault_in_safe_writeable - fault in an address range for writing
* @uaddr: start of address range
Commit a48b73eca4ce ("btrfs: fix potential deadlock in the search
ioctl") addressed a lockdep warning by pre-faulting the user pages and
attempting the copy_to_user_nofault() in an infinite loop. On
architectures like arm64 with MTE, an access may fault within a page at
a location different from what fault_in_writeable() probed. Since the
sk_offset is rewound to the previous struct btrfs_ioctl_search_header
boundary, there is no guaranteed forward progress and search_ioctl() may
live-lock.
Use fault_in_subpage_writeable() instead of fault_in_writeable() to
ensure the permission is checked at the right granularity (smaller than
PAGE_SIZE).
Signed-off-by: Catalin Marinas <[email protected]>
Fixes: a48b73eca4ce ("btrfs: fix potential deadlock in the search ioctl")
Reported-by: Al Viro <[email protected]>
Cc: Chris Mason <[email protected]>
Cc: Josef Bacik <[email protected]>
Cc: David Sterba <[email protected]>
---
fs/btrfs/ioctl.c | 7 ++++++-
1 file changed, 6 insertions(+), 1 deletion(-)
diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c
index 238cee5b5254..d49e8254f823 100644
--- a/fs/btrfs/ioctl.c
+++ b/fs/btrfs/ioctl.c
@@ -2556,8 +2556,13 @@ static noinline int search_ioctl(struct inode *inode,
key.offset = sk->min_offset;
while (1) {
+ size_t len = *buf_size - sk_offset;
ret = -EFAULT;
- if (fault_in_writeable(ubuf + sk_offset, *buf_size - sk_offset))
+ /*
+ * Ensure that the whole user buffer is faulted in at sub-page
+ * granularity, otherwise the loop may live-lock.
+ */
+ if (fault_in_subpage_writeable(ubuf + sk_offset, len))
break;
ret = btrfs_search_forward(root, &key, path, sk->min_transid);
On Wed, Apr 06, 2022 at 07:09:22PM +0100, Catalin Marinas wrote:
> diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c
> index 238cee5b5254..d49e8254f823 100644
> --- a/fs/btrfs/ioctl.c
> +++ b/fs/btrfs/ioctl.c
> @@ -2556,8 +2556,13 @@ static noinline int search_ioctl(struct inode *inode,
> key.offset = sk->min_offset;
>
> while (1) {
> + size_t len = *buf_size - sk_offset;
> ret = -EFAULT;
> - if (fault_in_writeable(ubuf + sk_offset, *buf_size - sk_offset))
> + /*
> + * Ensure that the whole user buffer is faulted in at sub-page
> + * granularity, otherwise the loop may live-lock.
> + */
> + if (fault_in_subpage_writeable(ubuf + sk_offset, len))
> break;
This doesn't need a new 'len' variable. It's a left-over from the v2
where fault_in_writeable() took the size and a min_size argument, both
being 'len'.
--
Catalin
On Wed, Apr 06, 2022 at 07:09:22PM +0100, Catalin Marinas wrote:
> Commit a48b73eca4ce ("btrfs: fix potential deadlock in the search
> ioctl") addressed a lockdep warning by pre-faulting the user pages and
> attempting the copy_to_user_nofault() in an infinite loop. On
> architectures like arm64 with MTE, an access may fault within a page at
> a location different from what fault_in_writeable() probed. Since the
> sk_offset is rewound to the previous struct btrfs_ioctl_search_header
> boundary, there is no guaranteed forward progress and search_ioctl() may
> live-lock.
>
> Use fault_in_subpage_writeable() instead of fault_in_writeable() to
> ensure the permission is checked at the right granularity (smaller than
> PAGE_SIZE).
>
> Signed-off-by: Catalin Marinas <[email protected]>
> Fixes: a48b73eca4ce ("btrfs: fix potential deadlock in the search ioctl")
> Reported-by: Al Viro <[email protected]>
Acked-by: David Sterba <[email protected]>